Current Server Time: 21:37 (Central Europe)

#apertus IRC Channel Logs

2013/12/01

Timezone: UTC


22:15
troy_s
Oh god.
22:15
troy_s
4k. 4k. 4k.
22:15
troy_s
Junk bullshit numbers everywhere.
22:15
Bertl
yeah, 4k looks completely different than 2k :)
22:16
troy_s
Such idiocy.
22:16
troy_s
Using the mythical numbers of the sunglasses-cum-camera vendor, the Panavision Genesis mark one was a 6k camera.
22:17
troy_s
The _only_ 4k camera currently on the market is the Sony F65, and I am quite certain no troll in this channel has ever touched one.
22:18
troy_s
(With perhaps an exception for rexbron)
22:25
Bertl
hehe
22:43
se6astian
time for bed
22:44
se6astian
already later than usual for me :)
22:44
se6astian
good night!
22:44
se6astian
left the channel
23:04
dmj_nova
joined the channel
23:06
dmj_nova1
left the channel
00:28
Sasha_C
joined the channel
00:43
Sasha_C
left the channel
01:15
FergusL
dmj_nova: why that question about low light with higher frame rate ?
01:22
dmj_nova
FergusL: depending on what needed changing to get the higher frame rate, it might involve such a design tradeoff
01:22
dmj_nova
of course, it's also possible that the light-sensitive region is exactly the same and some other circuitry is just being clocked higher
01:23
Bertl
I doubt they changed anything except for the LVDS readout
01:23
Bertl
you can get the same framerate already for a window
01:30
FergusL
I guess it's still obivous to you that received light will be divided by two ?
01:30
Bertl
at twice the framerate at least :)
01:34
rexbron_
joined the channel
01:35
rexbron
left the channel
01:45
dmj_nova
FergusL: yes, when doubling framerate recieved light will be halved
01:46
dmj_nova
my concern was mainly that the image not be noisier when framerates were equal
01:46
dmj_nova
say at 24fps with 1/48th shutter speed
01:53
FergusL
I don't get it
02:23
dmj_nova
FergusL: what don't you get?
02:24
FergusL
what do you mean by equel ?
02:24
FergusL
what is equal ?
02:34
dmj_nova
so you have the old CMV12000 and the new one
02:35
dmj_nova
Instead of pushing the new one at higher framerate, you use both with the same settings
02:35
dmj_nova
then compare the images each sensor produces
02:36
dmj_nova
do you:
02:36
dmj_nova
a) get identical images
02:36
dmj_nova
b) the new sensor produces a noiser image
02:36
Bertl
only CMOSIS can answer that
02:36
dmj_nova
Bertl: I suspect so, but it's a good question to ask
02:36
dmj_nova
was just explaining what I meant to FergusL
02:36
Bertl
go ahead, ask them
02:40
Bertl
in the time you spent today asking people the question who cannot know the answer and explaining why you think it is a good question, you could have already formulated a proper request/questions for cmosis
02:55
dmj_nova
Bertl: true
02:56
dmj_nova
troy_s: I do know some who have used the F65.
02:57
dmj_nova
I think it was used for Tears of Steel
03:07
FergusL
no, only planed to be shot on F65, was shot on Red Epic... or maybe they shot the vfx shot with F65
03:07
FergusL
still, nobody here remotely approached one
03:07
FergusL
maybe me at the AFC micro-salon where Sony has a booth !
03:08
dmj_nova
FergusL: Tears of Steel was *all* vfx shots
03:08
FergusL
sorry I meant green screen shots
03:10
dmj_nova
again, all green screen shots :P
03:10
dmj_nova
http://mango.blender.org/production/4-tb-original-4k-footage-available-as-cc-by/
03:10
dmj_nova
not claiming anyone in the channel has used it
03:12
dmj_nova
though some in #blender and #blendercoders will have
03:13
FergusL
the bridge shots were shot 2 passes too ? plate + actors
03:15
dmj_nova
FergusL: nope, bridge shots were one take, you're right on that
03:15
dmj_nova
though they do have background plates for some shots
03:16
dmj_nova
everything but the bridge is green screen or cgi though
03:17
Bertl
off to bed now ... have fun!
03:17
FergusL
same, night
07:43
guesst
dmj_nova: depends on shutter mode, if you shot 1/48 at 24 or 48 fps, the lower framerate makes higher noise in GS, in RS the output will be equal
07:47
dmj_nova
guesst: :-/kinda tired of explaining what the original question was but here goes:
07:48
dmj_nova
so CMOSIS came out with a new version of the sensor that can do up to 300 fps.
07:48
guesst
it is not public info yet?
07:49
dmj_nova
I had asked on the list whether this new version had any adverse effects on noise at the slower framerates the older version could achieve
07:49
dmj_nova
our guess is that they're just clocking the same chip higher without any changes that affect image quality, but I should ask CMOSIS to know for sure
07:51
guesst
so for that change, my answer is that it can perform better in rolling shutter mode
07:52
guesst
well, in global too.. you just get the image out quicker and gets less corrupted
07:55
dmj_nova
there is no rolling mode to my knowledge
07:56
guesst
also note, that when you run the sensor at the max speeds, you might end up with disabling some onchip noise correction..
07:56
dmj_nova
also, how does being able to shoot 150fps vs 300fps change how things look at 24 or 48 fps?
07:56
guesst
so you have just GS, that is bad
07:56
guesst
the image gets noise, especially the bottom part
07:57
dmj_nova
well, it's just the way it is :)
07:57
guesst
so with 300fps readout you get just half the noise, over 150
07:57
guesst
does not need to be :) there are sensors capable both rs/gs
07:57
dmj_nova
guesst: but you collect half the light
07:57
dmj_nova
which means using more gain
07:58
guesst
no
07:58
guesst
with 24 fps you can collect up to 1/24
07:58
guesst
but you readout the global buffer at 150 vs 300
07:58
dmj_nova
sigh
07:58
dmj_nova
no, at 150 fps, you can only have a shutter speed up to 1/150th of a second
07:59
guesst
the exposition and readout are two separete processes
07:59
guesst
and can happen at different speeds
07:59
guesst
readout >= exposition (fps)
07:59
guesst
or read time < expo time
08:00
dmj_nova
output frame interval >= exposure duration
08:01
dmj_nova
if there is a frame 150 times per second
08:01
guesst
ther is no
08:01
guesst
there is 1 frame 24 times
08:01
guesst
get out in speed of 1/150 s
08:01
guesst
vs 1/300s
08:01
dmj_nova
/facepalm
08:02
dmj_nova
think slow motion
08:02
dmj_nova
okay
08:02
dmj_nova
you shoot 150 or 300 frames every second
08:02
guesst
you shoot 24 :)
08:02
guesst
but for slowmo, there is of course loss of light
08:02
dmj_nova
the new sensor can shoot up to 300 frames every second
08:03
dmj_nova
the old one can shoot only 150 frames every second
08:03
guesst
that both applies for full 12MP res?
08:04
dmj_nova
my question was whether the new 300fps sensor would perform as well in low light at equivalent framerates and exposure times
08:04
dmj_nova
both sensors can shoot at lower framerates
08:04
guesst
so that is what i try to explain
08:05
guesst
in equal frame rate, shutter speed and gain setting, the new one could do better, given that the data receiving electronics is capable of higher speeds
08:06
dmj_nova
I'd actually guess that the electronics are exactly the same.
08:06
dmj_nova
and the only difference is that this is a new binning class
08:07
guesst
then you get no benefit, except for process variation which can make both less or more noise :)
08:07
dmj_nova
but I'm not sure if this is the case
08:07
guesst
binning like pixel merging? so you are not getting full res now? !
08:07
dmj_nova
no, not pixel merging
08:07
dmj_nova
binning is a term in semiconductor manufacturing
08:09
dmj_nova
rather than throw out chips that can't handle full clock speed or have a bad memory unit or processing core, you disable the bad components or clock them slower, then sell them as a lower model
08:09
dmj_nova
In the case of Sensors, they bin chips by the number and nature of bad pixels
08:09
guesst
i would say it is a different design
08:10
dmj_nova
so this is one of 3 things:
08:12
dmj_nova
1) they changed everything (including the light detecting region) -> could impact noise/sensitivity
08:12
dmj_nova
2) they only changed some circuitry relating to the signal transmission to handle higher clock speed -> unlikely to affect image quality
08:12
dmj_nova
3) this is a higher bin but otherwise identical -> almost certainly equal or slightly better image quality
08:13
guesst
if you can get the new sensor, you can compare it
08:13
dmj_nova
or just ask the company what they changed
08:13
guesst
are you doing some comparison agains red/arri anyway, before going to next step
08:14
dmj_nova
we couldn't compare until our next steps anyway
08:14
guesst
nothing has more value than a direct comparison :) you could not tell from the changed items if it is better or worse
08:14
dmj_nova
we're still at first images
08:14
guesst
so what is next?
08:15
dmj_nova
more advanced prototype. We need to design a processing board before we can get full speed video output
08:15
guesst
i am thinking so
08:16
dmj_nova
plus our initial images aren't calibrated or anything
08:17
dmj_nova
imagine RED footage without the dark noise calibration.
08:17
guesst
so the funding campaign is moving to next year? what do you need for the processing board?
08:17
dmj_nova
And our first sensor was defective (and therefore cheaper)
08:19
dmj_nova
so a sensor that did pass QA is necessary to test image quality too
08:19
dmj_nova
anyway, I gotta get back to work
08:20
dmj_nova
spent way to much time explaining this stupid question :P
08:20
guesst
yeah, just borrow old/new sensor of production quality and compare them yourself
09:04
philippej
joined the channel
09:32
dmj_nova
philippej: good point on the lvds lines
09:32
philippej
we need to summarize all discussiobs somewhere else
09:32
philippej
it's too time consuming to follow up with everything
09:40
philippej
left the channel
09:48
troy_s
dmj_nova: No big thrill here. ;)
09:48
dmj_nova
what thrill?
09:48
troy_s
ToS was all F65, but boy was post porked.
09:50
troy_s
The overall FPS has no impact.
09:50
troy_s
(aside from the higher FPS)
09:51
troy_s
dmj_nova: Of being around an F65.
09:51
troy_s
Not exactly thrilling here.
09:52
dmj_nova
post porked?
09:52
dmj_nova
You mean that most everything is CGI?
09:57
troy_s
I mean the pipeline was... a little odd.
09:58
dmj_nova
ah, yeah I recall some of that
09:59
dmj_nova
format conversions, funny colorspaces and such
09:59
troy_s
At the time it still hadn't set in what I had been harping on about primaries and such.
09:59
troy_s
There is so much of a knowledge gap regarding color etc.
09:59
troy_s
And scene referred versus display referred etc.
09:59
dmj_nova
yeah
10:00
troy_s
There are _still_ some folks that would like to insist that “linear” is a color space.
10:00
troy_s
Which drives me bloody batty.
10:00
dmj_nova
hehe
10:01
dmj_nova
though to be fair, linear is about all 3D rendering can or should care about
10:01
dmj_nova
the algorithms kinda need it to be that way
10:01
dmj_nova
of course what you do from that linear output is different
10:02
troy_s
I remember when they were sitting on the F65 footage and didn't know what to do with it.
10:02
troy_s
I demoed how to process it via OCIO.
10:02
troy_s
Linear is vastly more important than that.
10:03
troy_s
Both display linear and scene linear is the _only_ way to correctly blend light.
10:03
dmj_nova
yes, that's what I meant
10:03
troy_s
So even a dissolve is screwed if you aren't in a display or scene linear model.
10:03
dmj_nova
You need linear to "add" light
10:03
troy_s
Add, blend, multiply, anything really.
10:04
troy_s
Aside from a timed dissolve that appears perceptually timed.
10:05
philippej
joined the channel
10:05
troy_s
But the linear side (either scene or display) is only one half of the issue. The other half - the primaries (or color essentially) is equally important.
10:06
troy_s
And all of this is inevitably crap that people have to learn on the road to attempting what appeared so simple when they started (programming a decoder, player, etc.)
10:07
troy_s
It is always why I giggle and sigh when I read people wanting to add “simple grading” to an application and yet they haven't even begun to figure out color.
10:09
dmj_nova
sounds like something that should be documented for such people
10:10
troy_s
It is a path of pain and nightmare.
10:10
troy_s
Not worth it. No one would ever attempt it. Hence why I chuckle at some of the libraries.
10:11
troy_s
Prudent to stick to the industrial ones and then figure out why.
10:12
dmj_nova
no one would ever attempt documenting it?
10:13
troy_s
Most of the time the information is out there. Many are often stubborn or unwilling.
10:13
troy_s
Which can be positive as well I suppose.
10:14
troy_s
Selan's document is wonderful.
10:14
dmj_nova
link?
10:19
guesst
so is apertus more about hardware or software?
10:25
troy_s
guesst: A camera is both.
10:25
troy_s
Unavoidable byproduct.
10:25
dmj_nova
you can't build a functional piece of hardware without software to make it run
10:25
troy_s
dmj_nova: http://cinematiccolor.com
10:25
philippej
left the channel
10:26
dmj_nova
we're building a hardware device, but software is needed to make it do stuff
10:27
guesst
so the primary goal is to make a camera, and the rest is just to make use of it
10:27
guesst
i was thinking for a moment that you are doing software as primary and a camera is unavoidable as so far there are none to use
10:30
dmj_nova
no, the project is about making the camera
10:31
dmj_nova
there may be software projects out there for with the axiom is beneficial because there aren't equivalents, but it's apertus' business to develop the thing
10:32
troy_s
guesst: Not sure I would divide the goal into a metric of primary / secondary. The model is broken
10:32
troy_s
A functional camera consists of software and hardware. Unavoidable.
10:33
troy_s
Simply parts of a whole
10:33
guesst
i meant camera(hw/sw) vs. digital cinema toolchain
10:33
troy_s
guesst: Well processing images has to take the pipeline into account.
10:34
troy_s
Again, unavoidable. Or rather, if you avoid / ignore it, the byproduct is a sub-optimal design often (audience depending)
10:36
guesst
for me it seems already suboptimal by trying to cover all those aspects
10:38
dmj_nova
if you don't bother to characterize your colorspace you'll be unsuitable for many uses
10:43
dmj_nova
huh, this chart has suddenly helped me understand why I grade images the way I do.
10:43
dmj_nova
S-curve
10:47
Bertl
morning everyone!
10:57
dmj_nova
troy_s: that article is great
11:02
Sasha_C
joined the channel
11:35
se6astian
joined the channel
11:36
se6astian
hello!
11:36
Sasha_C
Hi Se6astian. What's news?
11:37
se6astian
not much I guess :)
11:38
Sasha_C
I've been reading up on the logs in our archives. I didn't realise so much stuff was happening whilst I was asleep :(
11:42
se6astian
:)
11:44
dmj_nova
se6astian: so that nofilmschool article seemed not to understand what it was talking about
11:46
Sasha_C
dmj_nova: That wouldn't be the first time this has happened ;)
11:46
Sasha_C
left the channel
11:46
dmj_nova
thinking that a camera that is likely limited by processing or storage limits will become faster based on a sensor upgrade
11:46
Sasha_C
joined the channel
11:47
dmj_nova
Sasha_C: thinking that a camera that is likely limited by processing or storage limits will become faster based on a sensor upgrade
11:49
guesst
they just might lower the noise and make something with the black spots in brigth areas :)
11:50
guesst
dynamic range of 60dB equals to 10bit / 10stops, right?
11:50
Bertl
the 'black sun' artefacts
11:51
guesst
yes
11:51
guesst
was looking on IBC for them.. they got it :P
11:52
dmj_nova
'black sun' artifacts?
11:54
guesst
when you apply extremely high light input to a sensor, it corrupts the referenced value and the conversion leads to usually dark / black pixel, but can be random
11:55
guesst
http://vimeo.com/52965171
11:56
guesst
anyway it is not much adviced to point it to sun.. you might vaporize the color filter :)
11:56
guesst
but on ibc we turned the camera up to the ceiling - to sharp lights and it was there too
12:02
se6astian
most sensors have settings related to dark spot sun supressions
12:02
se6astian
so does the cmv12000
12:02
se6astian
anyway, gotta go :)
12:02
se6astian
see you later
12:02
se6astian
left the channel
12:02
Sasha_C
caio
12:48
Sasha_C
left the channel
13:17
Bertl
test 1 2 3
13:32
FergusL
Bertl we copy, ove
13:32
FergusL
r
13:49
Bertl
hehe
14:01
ThatCantBe
left the channel
15:08
Bertl
off for a nap ... bbl
15:27
se6astian
joined the channel
15:29
se6astian
good evening
17:24
ThatCantBe
joined the channel
18:37
Bertl
back now ...
19:11
philippej
joined the channel
19:22
philippej
Hi everyone
19:53
FergusL
'evening there
19:53
philippej
left the channel
21:29
se6astian
good night!
21:29
se6astian
left the channel
22:02
troy_s
guesst: Sub optimal to cover them? Yes if you think building software. But you need to think harder.
22:02
troy_s
guesst: Things like codecs for editorial, file formats for online, etc. are all dreadfully imoortant.
22:03
troy_s
guesst: _Every_ design choice is non-neutral and comes with very particular implications for worlflows.
22:06
theverant
joined the channel
22:08
Sasha_C
joined the channel